6.5% of GDP in France: this figure which proves that the French delay in AI is not what we think

Why, when between 2020 and 2024, researchers based in France published each year a number of research articles in AI per million inhabitants equivalent to more than three quarters of the American level, does hourly productivity in France continue to mark time and decline compared to American productivity? A luminous note by Antonin Bergeaud and Simon Bunel, entitled “Why AI does not (yet) make France more productive”, gives us the keys to the enigma. In summary: the dropout is explained neither by a lesser investment effort nor by a deficit in research, but by an insufficient penetration of AI within the organization of companies.

In 2025, IT spending will represent 6.5% of GDP in France, a level close to that of the United States (6.8%). And France is within the European average in terms of AI research. If the delay in investment in information and communication technologies (ICT) has long been put forward to explain the productivity gap between France and the United States, this explanation no longer seems to prevail for around fifteen years.

Anthropic is targeting a valuation of up to $1.4 trillion for its IPO. The ability of companies to take advantage of digital equipment and convert it into productivity gains is decisive. However, this catch-up in terms of IT investment was accompanied by a period of low productivity gains. Even more, the productivity gap compared to the United States has been concentrated in sectors where digital technologies are likely to produce the greatest gains. This observation invites us to shift the analysis: beyond digital equipment, it is the capacity of companies to take advantage of it and convert it into productivity gains which appears decisive.

That the obstacles to technological adoption are linked to skills and the organization of companies was already reflected in the fact that with comparable equipment, the subsidiaries of American multinationals established in Europe derived more productivity from information technologies than European companies: part of this advantage was attributable to their managerial practices, closer to those used in the United States. However, Bergeaud and Bunel show that the same phenomenon is reproduced with AI: French companies are struggling to achieve productivity gains. Only 18% of French companies use AI and, within these companies, only 28% of employees use it, compared to 43% in the United States.

The portion of their working time devoted to these tools is twice as low. Nearly seven out of ten French workers do not benefit from any encouragement, tools or training in AI from their employer. Above all, the authors show that the use of AI in French companies remains confined to targeted applications. Companies that have adopted AI use it as much as American companies for product development, but much less for internal processes, marketing, customer relations and human resources.

However, these cross-functional functions offer significant potential for productivity gains, but require rethinking the organization of work and better sharing of data. “Augmented professions”: what are these highly sought-after strategic professions that benefit from the rise of AI? The obstacles to the penetration of AI within companies are above all organizational.

Nearly seven in ten French workers (69%) do not benefit from any encouragement, tools or training in AI from their employer, compared to 44% in the United States. Furthermore, restructuring costs five times more in France than in the United States. If the costs linked to the use of data constitute a barrier for all European companies, neither the skills gap nor the obstacles declared by companies are sufficient to explain a significant part of the French delay.

Hence the importance of more assertive intervention by public authorities to invest in computing power – more than 50% lower than the European average per capita – and in the diffusion of AI in our productive and industrial fabric, as we recommended more than two years ago in our report “AI: an ambition for France”. France has many assets – cutting-edge researchers, abundant savings, carbon-free electrical energy thanks to nuclear power – so as not to miss the AI ​​train as it missed the ICT train. However, this absolute priority that must be given to the AI ​​revolution and to the policies of education, support for innovation and organization of the labor market, necessary to support it and get the most out of it in terms of growth and jobs, is surprisingly absent from the presidential election debate.

Machine-translated from French